key: cord-287751-52e0tlcu
authors: Dai, Qili; Ding, Jing; Song, Congbo; Liu, Baoshuang; Bi, Xiaohui; Wu, Jianhui; Zhang, Yufen; Feng, Yinchang; Hopke, Philip K.
title: Changes in source contributions to particle number concentrations after the COVID-19 outbreak: Insights from a dispersion normalized PMF
date: 2020-11-06
journal: Sci Total Environ
DOI: 10.1016/j.scitotenv.2020.143548
sha: 
doc_id: 287751
cord_uid: 52e0tlcu

Factor analysis models use the covariance of measured variables to identify and apportion sources. These models, particularly positive matrix factorization (PMF), have been extensively used for analyzing particle number concentrations (PNCs) datasets. However, the variation of observed PNCs and particle size distribution are driven by both the source emission rates and atmospheric dispersion as well as chemical and physical transformation processes. This variation in the observation data caused by meteorologically induced dilution reduces the ability to obtain accurate source apportionment results. To reduce the influence of dilution on quantitative source estimates, a methodology for improving the accuracy of source apportionment results by incorporating a measure of dispersion, the ventilation coefficient, into the PMF analysis (called dispersion normalized PMF, DN-PMF) was applied to a PNC dataset measured from a field campaign that includes the Spring Festival event and the start of the COVID-19 lockdown in Tianjin, China. The data also included pollutant gases and hourly PM2.5 compositional data. Eight factors were resolved and interpreted as municipal incinerator, traffic nucleation, secondary inorganic aerosol (SIA), traffic emission, photonucleation, coal combustion, residential heating and festival emission. The DN-PMF enhanced the diel patterns of photonucleation and the two traffic factors by enlarging the differences between daytime peak values and nighttime concentrations. The municipal incinerator plant, traffic emissions, and coal combustion have cleaner and more clearly defined directionalities after dispersion normalization. Thus, dispersion normalized PMF is capable of enhancing the source emission patterns. After the COVID-19 lockdown began, PNC of traffic nucleation and traffic emission decreased by 41% and 44%, respectively, while photonucleation produced more particles likely due to the reduction in the condensation sink. The significant changes in source emissions indicate a substantially reduced traffic volume after the implement of lockdown measures.

Positive associations between ambient particulate matter (PM) pollution and subsequent detrimental health outcomes have been well documented in many epidemiologic studies (Song et al., 2017; Yin et al., 2019) . Given the severe fine PM (PM 2.5 , with aerodynamic diameter less than 2.5 µm) pollution in China, PM 2.5 was ranked as the first leading mortality risk factor in 2015 (Cohen et al., 2018) .

Compared with larger sized particles with higher mass concentrations, smaller particles have lower mass concentrations but higher particle number concentrations (PNCs) and larger surface areas that are more likely resulting in increased adverse health effects (Meng et al., 2013; Yin et al., 2019) . A recent time-series analysis in Northern China (Shenyang) reported that short-term exposure to increased PNCs for particles with size less than 0.5 µm were significantly associated with total and cardiovascular mortality (Meng et al., 2013) . They found that these associations between mortality and PNCs tended to be independent of particle mass concentrations and simultaneous exposures to gaseous pollutants. Previous studies also highlighted that the limited availability of size-resolved PNC measurements prevented obtaining adequate epidemiologic evidence regarding associations with PNCs (Ibald-Mulli et al., globally (Isaifan, 2020; Venter et al., 2020) . For example, air quality in China has improved with the reduction of non-essential industrial and motor vehicle usage as reported by Liu et al. (2020) , who has investigated the spatial and temporal characteristics of nighttime light radiance and air quality index values before and during the pandemic in mainland China. The COVID-19 outbreak provides an opportunity to evaluate the public health benefits of air quality improvement. A recent study found that COVID-19 forced-industrial and anthropogenic activities lockdown are likely have saved more lives by preventing ambient air pollution than by preventing infection . Their cross-sectional study conducted in the United States indicated that an increase of only 1 μg m -3 in PM 2.5 was associated with an 8% increase in the COVID-19 death rate. Thus, an emphasis is needed on the importance of continuing to enforce existing air pollution regulations to protect human health both during and after the COVID-19 crisis. The environmental impact of the pandemic is of particular interest to governments and the public since it is crucial for developing post-pandemic pollution control strategies. However, there are no reports to date on changes in source contributions to health risk relevant PNCs during the outbreak of COVID-19.

To identify particle sources and calculate their contributions, source apportionment studies have been conducted to support control strategies that can further reduce the health burden of PM. Positive matrix factorization (PMF) has been widely applied to 6 et al., 2011; Lorelei de Jesus et al., 2020) . Dilution is recognized as a crucial process that induces other processes to alter the particle number and size distributions (Gidhagen et al., 2005; Jacobson and Seinfeld, 2004; Ketzel and Berkowicz, 2004) . Kumar et al. (2011) reviewed the dynamics and dispersion modelling of nanoparticles from road traffic in the urban atmospheric environment and concluded that dilution is the most important process that needed to be considered with highest priority in dispersion models irrespective of any spatial scales. Freshly emitted particles and new particles formed via nucleation process are diluted in the air and undergone transformation processes. The conventional use of PMF to apportion PNCs is to extract size distribution profile from observed particle size distributional data on the basis of the internal covariance of particles from different detected sizes. Due to the variation in dispersion, some of the information content in the observation data will be certainly lost. To investigate the changes in source contributions to PNCs after the outbreak of COVID-19, a newly proposed dispersion normalized PMF (DN-PMF) that incorporated the ventilation coefficient (VC) into the PMF analysis aimed of reducing meteorological influence on the analysis was applied to a PNCs dataset measured from January 15, 2020 to February 13, 2020, included the start of the COVID-19 lockdown in Tianjin, China. The database also included pollutant gases and hourly PM 2.5 speciated composition data. suburban site surrounded by several universities, and distant from major highways and high traffic zones. A suite of instruments was operated in the supersite building with sampling inlets on the roof terrace.

Particle number concentrations with sizes ranging from 7.2 nm to 778 nm were measured using a universal scanning mobility particle sizer spectrometer (U-SMPS) (PALAS, Germany). Other air pollutants, including PM 1 and PM 2.5 , sulfur dioxide (SO 2 ), nitrogen oxides (NO 2 ), carbon monoxide (CO) and ozone (O 3 ) were also recorded hourly. Organic carbon (OC) and element carbon (EC) were measured hourly using a semicontinuous thermal-optical carbon analyzer (Focused Photonics Inc., China) and water-soluble ions (SO 4 

where is dispersion normalized concentrations, is the concentration observed during period i, and is the average value of VC over the entire measurement campaign. In this case, is 582 m 2 s -1 . The concentrations and uncertainties of PNC dataset were normalized on a sample by sample basis. The dispersion normalized concentrations and uncertainties were used as the input matrix to the PMF analysis.

US EPA PMF 5.0 was used for PMF analyses. Details of PMF are available elsewhere (Hopke, 2016; Paatero, 1999) . In addition to measured PNC data, auxiliary variables include gaseous pollutants (SO 2 , NO 2 , CO and O 3 ), PM 1 , PM 1-2.5 (PM 2.5 -PM 1 ), radiation and selected potential source tracer species (OC, EC, NO 3 -analyses to support factor interpretation. Adding additional variables increases the numbers of edge points and thereby reduce rotational ambiguity (Emami and Hopke, 2017) .

All missing data values were replaced by the median value for the given size bin or species and the uncertainty was set to three times that value. Measurement errors for PNC were estimated as follows (Ogulei et al., 2006b; Squizzato et al., 2019) :

where is the estimated measurement error for size bin j of i th sample, Nij is the measured concentration for size bin j of i th sample, ̅ is the arithmetic mean of the measured concentration for size bin j. was commonly used as a constant throughout all size bins and pollutants ( =0.01). Given that the lowest and highest bins of PNC were reported with elevated measurement error (Liu et al., 2014; Wiedensohler et al., 2018) , we set higher uncertainties to these size bins. The first two lowest size bins and highest size bins were assigned with a multiplier of , and the second two lowest size bins and highest size bins were assigned with multiplier of . For subsequent lower and higher size bins, the multiplier was set with a decrement of 0.5 until it approaches unit. The overall uncertainty matrix was computed as (Ogulei et al., 2006a) :

where was estimated from Eq. 3 and is a constant. C 3 values from 0.01 to 0.5 were tested. A value of 0.1 was assigned to C 3 for all size bins and 0.15 to SO 2 , NO 2 , CO, O 3 , PM 1 , PM 1-2.5 and radiation, while C 3 was set as 0.2 for the total PNC to obtain an optimal solution in mathematical sense. For selected chemical species, these species concentrations below method detection limit (MDL) values were replaced by J o u r n a l P r e -p r o o f Journal Pre-proof half of the MDL value. The corresponding uncertainties of these values were set at five sixths of the MDL values (Polissar et al., 1998) . Missing chemical species values were estimated by linearly interpolating the closet non-missing values and the corresponding uncertainties were increased by a factor of 3. All auxiliary variables were assigned as "weak" and the total PNC was set as the "total variable" with uncertainty tripled. The F-value for total PNC was used to normalize the size bin data and factor contributions. The F-value for PM 2.5 was used to normalize the chemical 

The time series of PNC, PM 1 mass concentration and VC values are shown in Fig. 1 .

The hourly average total PNC from 7.2 to 778 nm over the measurement campaign was 14397 ±5695 pt cm -3 , ranged from 5490 to 48372 pt cm -3 . The average mass concentration of PM 1 was 52.2 ± 42.6 µg m -3 , ranged from 1.0 to 224.1 µg m -3 . In 

Solutions using PMF with four to nine factors were explored for the original data and dispersion normalized data. The best solution with the optimal number of factors was evaluated with selection criteria of appropriately narrow distributions of scaled residuals of PNCs and the physical interpretability of factors in terms of (a) examination of size factor profiles and its association with external variables, (b) source directionality from CBPF plots, and (c) diel patterns.

Eight-factor models were finally selected as the best solutions for both the S4 and 3, respectively. The factors are interpreted and discussed individually in the following section.

This factor presented three major size modes: a nucleation mode peaked at ~8-20 nm, a small sized mode peaked around 30-50 nm and an accumulation mode peaked at 100-200 nm (Fig. 3 ). On-line measurements of the size distribution of particles in flue J o u r n a l P r e -p r o o f and stack gas of a municipal waste incineration plant in Europe showed a bimodal size distribution with peak mode at 90-140 nm and a smaller mode at approximately 40 nm (Maguhn et al., 2003) . It was also found that particles tend to growth by absorption and coagulation after the boiler. The chemical species profile had high contributions of Cr and Cu with narrow DISP intervals (Fig. 3) , which are tracers of municipal waste incineration. Lu et al. (2018) reported abundant Cr in bottom ash from plastic materials incineration. The CBPF plot of incinerator also shows a very clear NW wind direction pointing to the municipal incineration plant. Thus, this factor was assigned to municipal incinerator. On average, municipal incinerator had the lowest contribution to total PNC during the campaign (3.2%, Fig. S5 ).

Traffic has been recognized as one of the major sources of PNC. Two traffic factors were identified as traffic nucleation and traffic emission with particle major size modes about 10-30 nm and 30-150 nm, respectively. As been previously reported in literature (Morawska et al., 2008; Vu et al., 2015) , diel pattern of the traffic nucleation factor has a sharp morning rush hour peak (6:00-8:00 am) and had enhanced contributions more broadly distributed beginning in the afternoon. While the traffic emission factor showed a strong peak in the morning after the rush hour (8:00-10:00 am) and a minor peak in the evening rush hour. Both traffic factors had high concentrations of NO 2 and explained 5-10% of the EC variation, indicating its traffic exhaust nature including diesel emissions .

The presence of some O 3 and radiation in the traffic emission factor, suggests that particles exhaust from vehicle emitted in the morning were possibly subjected to photochemistry processes.

The CBPF plot of traffic nucleation indicates its directionality corresponded to the roads located ~1.5 km and 3.0 km southwest of the measurement site (Fig. S1 ). The traffic emission factor has a stronger association with southeasterly winds and a predominant occurrence for wind speeds between 3-5 m s -1 . There are two major highways situated ~18 km east of the measurement site and a highway extending to the southeast. Both traffic factors are heavily affected by local traffic emissions.

The SIA (ammonium nitrate and sulfate) had three modes in the particle number size profiles with the major size mode between 180-600 nm, an Aitken mode and an ultrafine mode size ranges, which is similar with the observations such as in Rochester, NY (Squizzato et al., 2019) . The explained variations of particles in the accumulated mode increased with increasing size. The presence of high nitrate, sulfate and ammonium in the species profile support its assignment as secondary inorganic material. The wide particle size range of SIA factor suggests that it originated from both gas-to-particle conversion of local NO x and regional transportation of sulfate.

Some EC present in the species factor may be caused by the condensation of secondary material on freshly emitted EC particles. The morning peak of SIA at 9:00-10:00 am likely resulted from downmixing of transported secondary particles from aloft after the breakup of overnight inversions. The broad afternoon minimum was likely due to the mixing layer height dynamics. The CBPF plot of SIA shows higher probability values with moderate speed winds (~2 m s -1 ) come from downtown Tianjin (NW) more than other directions, highlighting the regional transport nature of SIA during heavy particulate pollution episodes. It was the largest contributor to PM mass concentration but accounted for only 6.2% of total PNC over the measurement campaign.

In addition to new particle formation from traffic emission with a dominant particle size ranged from ~10-30 nm as typically observed during rush hour as discussed above, a factor with particle size peaking in the nucleation mode size range (<20 nm), and characterized by high explained variations of O 3 and solar radiation with narrow DISP bands was identified as new particle formation (NPF) through photochemistry.

It shows a strong sharp increase around midday (12:00) concurrent with the highest solar intensities that drive photochemical processes. Photonucleation occurred primarily during clean days with low PM mass concentrations (Rivas et al., 2020) .

The low PM concentrations result in low condensation sinks and allow for NPF to occur (McMurry and Friedlander, 1979) . As suggested by the CBPF plot, this factor is associated primarily with relatively higher wind velocities for east-south-easterly winds and southerly winds. The CBPF plot of photonucleation overlaps with traffic nucleation in south-south-west direction, suggesting air masses from this direction J o u r n a l P r e -p r o o f likely facilitate NPF. Easterly winds often associated with relatively clean Bohai Bay air possibly favors the nucleation process because of its low condensation sink.

Interestingly, the non-local feature of photonucleation tends to indicate that the measurement site is likely well separated from the emission sources. On average, this factor contributed 7.8% of total PNC.

Coal combustion was reported as an important source of ambient particles with size ranged from nanoparticles to coarse particles, depending upon various factors such as coal type, combustion condition, dilution ratio and residence time, and pollution control device (Vu et al., 2015) . The formation of particles emission from coal combustion involved in complex chemistry as it will undergo vaporizationnucleation/condensation-growth processes (Carbone et al., 2010) . Lipsky et al. (2002) reported a nucleation peak mode around 10 nm in the stack plume of coal combustion.

Particles tend to shift to large sizes after release into ambient air experiencing dilution, cooling, and condensation/coagulation processes. An experimental measurement of pulverized coal combustion suggested that ultrafine particles exhibited a unimodal distribution under 1500 K with size peak shifting from 10.76 nm to 38.46 nm as time evolved (Gao et al., 2017) . There is a paper mill with hot steam supplied by pulverized coal boilers situated ~6.5 km SE from the measurement site (Fig. S1 ). The CBPF plot presents a clear SE direction linking this factor to the paper mill. Thus, this factor with particle sizes peaking around 30 nm was attributed to coal combustion.

The gaseous and species profiles corroborate the assignment of this factor to coal combustion since it had relatively high concentrations of NO 2 , EC, SO 4 2and NO 3 -, and explained some of the SO 2 , CO, and As variations. The diel pattern presents a J o u r n a l P r e -p r o o f Journal Pre-proof middle day peak as the particles can be transported by the afternoon sea breeze from Bohai Bay.

The factor has a size mode ranging from ~50 nm to 300 nm and peaked in the accumulation mode (~120 nm). It was identified as residential heating. Its size distribution is similar with particles emission from residential wood pellet boilers (Chandrasekaran et al., 2011; Wang et al., 2019) . It is associated with some SO 2 , NO 2 , and CO with tight DISP bands, and moderate loading of PM mass concentration. The chemical species profile also supports coal as it explained ~15% of Cland ~10% OC with small intervals in mass fractions. Cl is a common marker of coals in northern

China (Yu et al., 2013; Li et al. 2019 ).

The diel pattern for this factor presents a clear morning peak (7:00-8:00 am) during stable atmosphere conditions with low wind speeds. PNC of residential heating during nighttime appears to be higher than that during daytime. The CBPF plot of residential heating factor has an association with all wind directions and low wind speeds (<2 m s -1 ).

The festival emission factor is attributed largely on the basis of strong chemical association with OC, EC, NO 3 -, SO 4 2-locations (Joshi et al., 2019; Kong et al., 2015; Moreno et al., 2010; Tian et al., 2014) .

These species have relatively narrow DISP intervals in the species profiles supporting the inclusion of firework emissions in this factor. OC, EC and sulfate in the profile were also likely emitted from coal/wood burning in large bond fires to celebrate the festival as reported in Dai et al. (2020) . Fireworks were generally banned in urban areas and Tianjin implemented a no-fireworks policy this year. However, people still likely utilize fireworks during these holiday periods. Based on the CBPF plot, the origins of these particles were located in residential areas where coal/biomass was burned for heating/cooking during the Spring Festival. People were required to stay at home after the implement of lockdown measures to prevent the spread of COVID-19

and thus, more people were in residential areas throughout the day requiring more heating and cooking than under normal circumstances. Fireworks emissions were correlated with residential coal/biomass burning. It should be noted that this factor contributed significantly to PM mass concentration, particularly PM 1 in the period after the start of the Spring Festival (January 24, 2020).

Since the DN-PMF reduced the influence of dispersion, the CBPF plots derived from DN-PMF results tend to have better defined directionality than the regular PMF based plots. 

This measurement campaign included the unusual event of the Chinese Spring Festival (SF) and lockdown measures enforced to prevent the spread of COVID-19.

Unlike the common sources under the "business as usual" scenario prior to January 24, there was a change in the number of sources due to the fireworks displayed during the SF. Source emission strengths also changed after the COVID-19 outbreak as emergency responses were implemented beginning on January 25 in Tianjin. To J o u r n a l P r e -p r o o f 

To reduce the influence of meteorology on quantitative source estimates, dispersion normalized PMF incorporating data normalized with the ventilation coefficient into PMF analyses, was applied to a PNCs dataset measured from a field campaign that includes the Spring Festival event and the start of the COVID-19 outbreak in Tianjin,

China. In addition to PNCs data, other variables include solar radiation, gaseous pollutants, particle mass concentrations and chemical compositional data were measured and included to facilitate the interpretation of factors. The PNCs dataset combined with these additional variables were normalized by VC values and then subjected to PMF analysis. Eight factors were resolved and labelled as municipal incinerator, traffic nucleation, secondary inorganic aerosol (SIA), traffic emissions, photonucleation, coal combustion, residential heating and festival emission.

To examine the effects of reducing meteorological influence on estimated source patterns, the diel pattern and source directionality for each source derived from conventional PMF (unnormalized) and DN-PMF were compared. The DN-PMF enhanced the diel patterns of photonucleation and the two traffic factors by sharpening the differences between daytime peak values and nighttime concentrations.

After dispersion normalization, PMF yielded better defined directionality of sources.

The directionalities of municipal incinerator plant, traffic emission and coal 

Estimates and 25-year trends of the global burden of disease attributable to ambient air pollution: an analysis of data from the Global Burden of Diseases Study

Dispersion Normalized PMF Provides Insights into the Significant Changes in Source Contributions to PM 2.5 after the COVID-19 Outbreak

Effect of adding variables on rotational ambiguity in positive matrix factorization solutions

Source apportionment of ultrafine and fine particle concentrations in Brisbane

Measurement and numerical simulation of ultrafine particle size distribution in the early stage of high-sodium lignite combustion

Urban scale modeling of particle number concentration in Stockholm

PMF Analysis of Wide-Range Particle Size Spectra Collected on a Major Highway

Review of receptor modeling methods for source apportionment

Epidemiological evidence on health effects of ultrafine particles

The dramatic impact of Coronavirus outbreak on air quality: Has it saved as much as it has killed so far?

Evolution of nanoparticle size and mixing state near the point of emission

Simultaneous measurements of mass, chemical compositional and number characteristics of aerosol particles emitted during fireworks

Modelling the fate of ultrafine particles from exhaust pipe to rural background: an analysis of time scales for dilution, coagulation and deposition

Multi-plume aerosol dynamics and transport model for urban scale particle pollution

The impacts of firework burning at the Chinese Spring Festival on air quality: insights of tracers, source evolution and aging processes

Sources of humic-like substances (HULIS) in PM 2.5 in Beijing: receptor modeling approach

Effects of sampling conditions on the size distribution of fine particulate matter emitted from a pilot-scale pulverized-coal combustor

Spatiotemporal Patterns of COVID-19 Impact on Human Activities and Environment in China Using Nighttime Light and Air Quality Data. arXiv 2005

Source apportionment of urban fine particle number concentration during summertime in Beijing

Long-term trends in PM 2.5 mass and particle number concentrations in urban air: The impacts of mitigation measures and extreme events due to changing climates

Composition and source apportionment of heavy metals in bottom ash from a municipal solid waste incinerator in Beijing

On-line analysis of the size distribution of fine and ultrafine aerosol particles in flue and stack gas of a municipal waste incineration plant: Effects of dynamic process control measures and emission reduction devices

New particle formation in the presence of an aerosol

Size-Fractionated Particle Number Concentrations and Daily Mortality in a Chinese City

Ambient nano and ultrafine particles from motor vehicle emissions: Characteristics, ambient processing and implications on human exposure

Effect of fireworks events on urban background trace metal aerosol concentrations: Is the cocktail worth the show

Analysis of indoor particle size distributions in an occupied townhouse using positive matrix factorization

Source apportionment of Baltimore aerosol from combined size distribution and chemical composition data

The multilinear engine -A table-driven, least squares program for solving multilinear problems, including the n-way parallel factor analysis model

Epidemiological evidence on the health effects of ultrafine particles

Atmospheric aerosol over Alaska: 2. Elemental composition and sources

Health burden attributable to ambient PM 2.5 in China

Long-Term Changes of Source Apportioned Particle Number Concentrations in a Metropolitan Area of the Northeastern United States

Estimation of the direct and indirect impacts of fireworks on the physicochemical characteristics of atmospheric PM 10 and PM 2.5

PM 2.5 source apportionment during severe haze episodes in a Chinese megacity based on a 5-month period by using hourly species measurements: Explore how to better conduct PMF during haze episodes

Conditional bivariate probability function for source identification

COVID-19 lockdowns cause global air pollution declines with implications for public health risk. medRxiv

Review: Particle number size distributions from seven major sources and implications for source apportionment studies

Emissions from in-use residential wood pellet boilers and potential emissions savings using thermal storage

Exposure to air pollution and COVID-19 mortality in the United States: A nationwide cross-sectional study. medRxiv

Associations between size-fractionated particle number concentrations and COPD mortality in

Characterization and source apportionment of PM 2.5 in an urban environment in Beijing

Particle Size and Mixing State of Freshly Emitted Black Carbon from Different Combustion Sources in China

Methodology, Writing-Original draft preparation Jing Ding: Assisting in sample collection, Data curation Congbo Song: Result interpretation, Writing -Review & Editing Baoshuang Liu: Writing -Review & Editing Xiaohui Bi: Resources, Investigation Jianhui Wu: Sample collection and analysis Yufen Zhang: Writing -Review & Editing Yinchang Feng: Resources

Result interpretation, Writing -Review & Editing

This work was financially supported by the National Key R&D Program of China